An Azure service that provides natural language capabilities including sentiment analysis, entity extraction, and automated question answering.
Hey @Siviwe Boyisi, Vodacom, – sorry to hear your entity models have gone missing. Usually, the fastest way to see who did what is to check the Activity (audit) logs on your Language Studio resource in the Azure portal and/or in the Foundry management center. Here’s what you can try:
- Azure Portal Activity Log
• Sign into portal.azure.com and navigate to the Cognitive Services or Language resource that backs your Language Studio project.
• In the left menu under Monitoring, click Activity log.
• Filter by the time window when you noticed the models disappear, and by Operation (look for “Delete” operations on deployments or projects).
• Click into any suspicious entries to see which user or service principal invoked the delete.
- Enable Diagnostic Settings (if not already on)
• Still in your Language resource, go to Diagnostic settings.
• Add a setting to send “AllLogs” or “Audit” to a Log Analytics workspace or Storage account.
• Once you’ve got logs shipping, you can run Kusto queries in Log Analytics (e.g. filter by OperationName contains “DeleteDeployment” or “DeleteProject”).
- Language Studio / Foundry Management Center
• Open https://ai.azure.com and sign in.
• Select your project under “Keep building with Foundry.”
• Go to Management center and look for a History or Audit tab. That will list entity‐level actions (create, update, delete).
If you still don’t see why those entity models vanished, can you share a few more details?
• Which kind of project are you using (Custom named-entity-recognition, CLU, QnA, etc.)?
• What’s the name, region, and resource group of the underlying Azure Language/Cognitive Services resource?
• Roughly when (date/time range) you noticed the entities disappear?
• Are there any other contributors or service principals with write access?
• Have you tried re-enabling or viewing Diagnostic settings yet?
With that info we can confirm whether it was a user/API delete, a deployment rollback, or a potential service-side issue.